Electric vehicle task allocation method considering dual-network energy consumption

Abstract

The invention discloses an electric vehicle task allocation method considering dual-network energy consumption, and the method comprises the steps: obtaining road network traffic flow data, charging pile position information, historical charging efficiency, task target related data and the like, and carrying out the mathematical modeling of an electric vehicle task allocation problem, a target function with a vehicle tracking target number, a maximum driving speed, road network edge selection, charging pile area selection and charging time as decision variables and with effective tracking duration and vehicle energy consumption as targets is formed, and a distributed dynamic evolution algorithm based on Nash optimization is adopted. Under the condition of ensuring the longest effective tracking duration and the lowest vehicle energy consumption, the optimal task allocation scheme is calculated, the tracking target, the running speed and the charging pile resources of the electric vehicle are reasonably allocated, and the task requirements are efficiently met.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and particularly to an electric vehicle task allocation method considering the energy consumption of a dual network. Background Art

[0002] In recent years, under the promotion of the national carbon peak and carbon neutrality strategies, along with the support for new quality productivity industries, new energy technologies and industries have achieved explosive growth. Especially electric vehicles, with their environmental protection, energy conservation and strong acceleration performance, are increasingly favored by the market. In addition, electric vehicles also provide an ideal carrier for advanced autonomous driving technologies. However, the battery performance of electric vehicles and the charging efficiency of charging piles are still the main bottlenecks restricting their development. When an electric vehicle cluster equipped with an intelligent driving system executes intelligent tasks, it is necessary to consider the impact of the traffic environment on the energy consumption of electric vehicles, as well as the distribution and charging performance of the power supply network mainly composed of the charging pile network, to ensure that the intelligent electric vehicle cluster can complete the tasks within the specified time. Such problems are highly dynamic and random, and traditional static optimization algorithms are difficult to effectively solve.

[0003] Evolutionary computation is a class of intelligent algorithms based on natural evolution, which is widely used in the solution of complex optimization problems, such as automotive structure optimization, job shop scheduling and process optimization, etc. Through hundreds of iterations of optimization and evaluation, evolutionary computation can obtain high-quality solutions that meet the requirements. When the environment changes little, using an evolutionary algorithm for electric vehicle task allocation is a wise choice. However, in actual task scenarios (such as target tracking), the traffic flow conditions of the road network and the service capabilities of the charging pile network are often highly dynamic. Therefore, the electric vehicle cluster must dynamically adjust the task allocation scheme to meet the task requirements and reduce energy consumption. Therefore, the long-time calculation of traditional evolutionary algorithms is unacceptable for such dynamic tasks. How to efficiently complete the dynamic optimization of electric vehicle task allocation using evolutionary computation methods within a reasonable time is still an urgent problem to be solved. Summary of the Invention

[0004] In order to overcome the above-mentioned shortcomings and deficiencies of the prior art, the purpose of the present invention is to provide an electric vehicle task allocation method considering the energy consumption of a dual network, specifically an electric vehicle task allocation method based on a distributed dynamic evolutionary algorithm of Nash optimization. This method comprehensively considers the energy consumption factors of the dual network (road network and charging pile network) to improve the efficiency and effect of electric vehicle task allocation.

[0005] The purpose of the present invention is achieved by the following technical solutions:

[0006] An electric vehicle task allocation method considering the energy consumption of a dual network includes the following:

[0007] S1 Obtain the trajectory data of each target, obtain the historical traffic flow data of each edge of the road network, and obtain the location information and historical charging efficiency data of each charging pile node in the charging pile network;

[0008] S2 Based on the trajectory data of the target and the dual-network data, conduct mathematical modeling on the electric vehicle task allocation optimization problem, define a model with the vehicle tracking target number, maximum driving speed, road network edge number, charging pile area number, and charging time as decision variables, and establish an objective function with the effective tracking duration and energy consumption as the objectives;

[0009] S3 Adopt a distributed dynamic evolution algorithm based on Nash optimization, with the longest effective tracking duration and the lowest vehicle energy consumption as the optimization objectives, to obtain the optimal electric vehicle task allocation method. The electric vehicle task allocation method satisfies the maximum driving speed of the vehicle with the lowest tracking ability threshold, and at the same time allocates the nearest charging pile to the electric vehicle that needs to be charged.

[0010] Further, in S2, based on the trajectory data of the target and the dual-network data, conduct mathematical modeling on the electric vehicle task allocation optimization problem, specifically as follows:

[0011] Use the LSTM model to model the traffic flow of the road network, train it with the historical traffic flow data of each edge of the road network, obtain a traffic flow time series prediction model for predicting the traffic flow of each edge of the road network in the next time step, and predict the traffic flow of each edge of the road network in the next time step;

[0012] Use the LSTM model to model the charging efficiency of the charging pile network, train it with the historical charging efficiency data, obtain an efficiency time series prediction model for predicting the charging efficiency of each charging pile area in the next time step, and predict the charging efficiency of each charging pile area in the next time step;

[0013] Based on the Kalman filter, model the target trajectory, train it with the target historical trajectory, obtain a target tracking time series prediction model, and predict the position prediction information of the target in the next time step.

[0014] Further, adopt a distributed dynamic evolution algorithm based on Nash optimization, with the longest effective tracking duration and the lowest vehicle energy consumption as the optimization objectives, to obtain the optimal electric vehicle task allocation method, specifically as follows:

[0015] S31 For each electric vehicle, calculate the objective function method as follows:

[0016] J(X i (k),{X j≠i (k)},u i [k:k+N-1],{u j≠i[k:k + N - 1]}, {E[k:k + N - 1]}); Among them, the objective function includes two sub-objective functions, namely the effective tracking duration and the vehicle energy consumption. J is the weighted average of the two sub-objective functions of the effective tracking duration and the vehicle energy consumption after normalization respectively;

[0017] k represents the current time slice number, N represents the next N time slices to be considered, X i (k) represents the state of electric vehicle i at time step k, {X j≠i (k)} represents the states of other electric vehicles in the system at time step k, u i [k:k + N - 1] represents the task assignment plan of electric vehicle i from time slice k to k + N - 1, {u j≠i [k:k + N - 1]} represents the task assignment plans of other electric vehicles from time slice k to k + N - 1, {E[k:k + N - 1]} represents the state of the environment from time slice k to k + N - 1;

[0018] The description of the sub-objective function of the effective tracking duration to be minimized is as follows:

[0019] Among them, N represents the number of time steps, T represents the number of target objects, e t is a Boolean function. If there is an electric vehicle performing a tracking task within the circle with the target object e t as the center and radius r at time step t, then e t is 1, otherwise it is 0;

[0020] The description of the sub-objective function of the vehicle energy consumption to be minimized is as follows:

[0021] Among them, ΔE i,k is the energy consumption of electric vehicle i at time step k, M is the number of unmanned vehicles, ΔE max is the possible maximum energy consumption of all unmanned vehicles in one time step, E max is the energy when each unmanned vehicle is fully charged, E i,k+N-1 is the energy of the unmanned vehicle at time step k. Since calculating this objective function involves multiple future time steps, the decision variables need to be brought into the above double-network model to predict the energy consumption situation;

[0022] For each electric vehicle i, with {u j≠i [k:k + N - 1]} fixed, use the genetic algorithm to iteratively optimize u i [k:k + N - 1], and then record and broadcast the optimal individual;

[0023] Each electric vehicle maintains a population, and the individuals in the population are isomorphic. The dimensions of an individual are the tracking target number, maximum driving speed, road network edge number, charging pile area number, and charging time of the electric vehicle at time step k;

[0024] S34 Each electric vehicle initializes the population and randomly selects an individual to broadcast;

[0025] S35 For each electric vehicle i, when fixing {u j≠i [k:k + N - 1]}, the genetic algorithm is used to iteratively optimize u i [k:k + N - 1], and then the optimal individual is recorded and broadcast;

[0026] S36 Each electric vehicle determines whether the Nash equilibrium is reached. If the Nash equilibrium is reached, each electric vehicle stores the optimal solution in the database and executes tasks according to this solution. The optimal solution is the electric vehicle task allocation method. Otherwise, it returns to S35 for further optimization.

[0027] Furthermore, the Nash equilibrium judgment formula:

[0028]

[0029] where M is all the intelligent electric vehicles in the system, is the optimal individual of electric vehicle i in the l-th iteration, is the optimal individual of electric vehicle i in all historical iterations, and ε n is a positive number, serving as a threshold.

[0030] Furthermore, it also includes that when the electric vehicle executes tasks, it monitors the vehicle state, target trajectory, road network traffic flow situation, and the performance of the sensor network in real time. Then, every once in a while, it calculates the fitness value of the current decision using the updated data. If the fitness value is lower than the threshold, the electric vehicle that recognizes that the solution has degenerated will broadcast this information, and then the system will perform dynamic optimization operations.

[0031] Furthermore, the dynamic optimization operation is specifically:

[0032] Increase the mutation probability of the genetic algorithm;

[0033] Shift the solution to the left: The optimal solution before shifting, that is, remove the task allocation of the executed time steps, use the time step corresponding to the current moment as the task allocation of the first time slice of this plan, and then shift it to the left in turn. Finally, fill the last valid time step into the remaining part. Then compare the shifted plan with the original optimal plan and retain the better one of the two, which is called the adjusted solution;

[0034] Re-initialize the population, then calculate the fitness values, and replace the individual with the worst fitness value in the population with the adjusted solution;

[0035] Use the database to guide the search: when the fitness is calculated, randomly select several individuals from the database, and at the same time select the corresponding number of the worst individuals in the current population, then compare the fitness values in turn. If the fitness value of the database individual is better than that in the population, replace the individual in the population with the database individual.

[0036] Furthermore, S36 further includes: when the database capacity is not full, directly add it to the database, otherwise replace the individual with the worst diversity in the database with the current individual.

[0037] Furthermore, the diversity calculation formula is:

[0038]

[0039] where C i,d is the value of the dimension d of individual i, N C is the number of individuals in the database, center d is used to represent the center of the database individuals, dc i is the distance of a certain individual from the center, which is used to characterize the diversity of the individual;

[0040] Furthermore, the state of the electric vehicle at time step k includes the position and remaining power of the electric vehicle.

[0041] Furthermore, the state of the environment includes the target position, road network state, and charging pile network state. The road network state includes the traffic flow, and the charging pile network state includes the charging efficiency and load condition.

[0042] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0043] The electric vehicle task allocation method considering the energy consumption of the dual networks based on the Nash optimization-based distributed dynamic evolution algorithm provided by the present invention mathematically models the electric vehicle task allocation optimization problem based on the collected information such as the target trajectory and dual network data, and respectively obtains the decision variables with variables such as the vehicle tracking target number, the maximum driving speed of the vehicle, the charging pile node selected by the vehicle, and the charging time, and the objective function with the effective tracking duration and the vehicle energy consumption as the objectives. By using the dynamic optimization method, the fast convergence ability of the algorithm in the dynamic optimization environment is improved, and the problem of poor real-time performance in the prior art when solving the dynamic optimization problem of electric vehicle task allocation considering the energy consumption of the dual networks is solved.

[0044] The electric vehicle task allocation method considering the energy consumption of dual networks based on the Nash optimization-based distributed dynamic evolution algorithm provided by the embodiments of the present invention uses LSTM, Kalman filtering, and model predictive control as aids, constructs a prediction model using information such as the target trajectory and dual network data, and uses model predictive control to consider long-term task allocation, so as to make an advance judgment on the future environment and improve the robustness of task allocation.

[0045] The electric vehicle task allocation method considering the energy consumption of dual networks based on the Nash optimization-based distributed dynamic evolution algorithm of the present invention introduces the Nash optimization algorithm on the basis of using distributed model predictive control. Each intelligent node (such as an electric vehicle) only considers the optimization of its own task allocation, thereby reducing the search space, improving the search efficiency, and at the same time, Nash optimization reduces the optimization conflicts between intelligent nodes and improves the search ability of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a flowchart of an electric vehicle task method considering the energy consumption of dual networks according to the present invention;

[0047] Figure 2 It is a flowchart of the dynamic optimization method of the present invention;

[0048] Figure 3 It is a schematic diagram of decision variables based on time series according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The following will further describe the present invention in detail in conjunction with embodiments, but the embodiments of the present invention are not limited thereto.

[0050] Embodiment

[0051] The embodiments of the present invention are implemented based on the execution of target tracking tasks by an intelligent electric vehicle cluster. Refer to Figure 1 , this embodiment provides an electric vehicle task allocation method considering the energy consumption of dual networks. The dual networks refer to the road network and the charging pile network, and include the following steps:

[0052] S1 Obtain the trajectory data of each target, obtain the historical traffic flow data of each edge of the road network, and obtain the position information and historical charging efficiency data of each charging pile node of the charging pile network;

[0053] S2 Based on the trajectory data of the target and the dual network data, perform mathematical modeling on the electric vehicle task allocation optimization problem, define a model with the vehicle tracking target number, maximum driving speed, road network edge number, charging pile node number, and charging time as decision variables, and establish an objective function with the effective tracking duration and energy consumption as the objectives, as shown in Figure 3 ;

[0054] Further explanation:

[0055] Use the LSTM model to model the traffic flow of the road network, and train it with the historical traffic flow data of each edge of the road network to obtain a traffic flow time series prediction model for predicting the traffic flow of each edge of the road network. The input and output of this model are as follows:

[0056] Input: The traffic flow of each edge of the road network at each time step. For example, at time step t, the input vector x t =[x t,1 , x t,2 , …, x t,N , where x t,i represents the traffic flow of the i-th edge at time step t) and the current time, where the current time includes the day of the week and the current hour;

[0057] Output: The traffic flow of each edge of the road network at the next time step.

[0058] Use the LSTM model to model the charging efficiency of the charging pile network, and train it with historical charging efficiency data to obtain an efficiency time series prediction model for predicting the charging pile efficiency. The input and output of this model are as follows:

[0059] Input: The charging efficiency of each charging pile node at each time step (for example, at time step t, the input vector x t =[x t,1 , x t,2 , …, x t,N , where x t,i represents the charging efficiency of the i-th charging pile node at time step t) and the load situation of each charging pile node at each time step, where the load situation includes the number of charging vehicles / capacity for charging vehicles.

[0060] Output: The charging efficiency of each charging pile node at the next time step.

[0061] Model the target trajectory based on the Kalman filter, and train it with the target's historical trajectory to obtain a target tracking time series prediction model. The input and output of the model are as follows:

[0062] Input: The historical motion trajectory of the target object, for example, [(x1, y1), (x2, y2), …, (x n , y n )], that is, the position information of the target object from time step 1 to time step n.

[0063] Output: The position prediction information of the target object at the next time step.

[0064] S3 adopts a distributed dynamic evolution algorithm based on Nash optimization, with the optimization goal of maximizing the effective tracking duration and minimizing the vehicle energy consumption, to obtain the optimal electric vehicle task allocation method. The electric vehicle task allocation method satisfies the maximum driving speed of the vehicle with the lowest tracking ability threshold, and at the same time allocates the nearest charging pile node to the electric vehicle that needs to be charged.

[0065] Further explanation: The distributed dynamic evolution algorithm based on Nash optimization evolves the objective function, and the decision variables include the vehicle tracking target number, the maximum driving speed of the vehicle, which edge of the road network the vehicle selects, the charging pile node and charging time selected by the vehicle, etc.

[0066] The optimization goals of the objective function are as follows:

[0067] Effective tracking time: If there is an electric vehicle performing a task within r meters around a certain target object, then this target is considered to be effectively tracked. The optimization goal is to maximize this tracking time. However, when calculating the objective function, we cannot directly obtain the decision result, so we need to predict the road network congestion situation and the future trajectory of this target object. At the same time, since the electric vehicle needs to be charged, we also need to predict the charging efficiency of each charging area to reasonably arrange the vehicles that need to be charged, so as to avoid the situation that all electric vehicles run out of power and the target object cannot be tracked;

[0068] Vehicle energy consumption: This is to ensure that the overall energy consumption is not excessive. This requires dispatching electric vehicles closer to the target object for tracking, requiring them to take smoother roads, and reasonably arranging the charging time, etc.

[0069] In the embodiment of the present invention, a distributed dynamic evolution algorithm based on Nash optimization is used to find a better task allocation scheme through iterative search and population evolution. At the same time, when the environment changes, the algorithm can quickly converge to quickly obtain a new task allocation scheme.

[0070] The present invention models the efficiency of road network traffic flow and charging pile network based on LSTM, and models the target trajectory based on Kalman filter. Then, model predictive control is used to model the optimization objective. This modeling method converts the decision variables into decision variables based on time series, and not only considers the allocation of tracking targets at the current time step, but also considers the target allocation in the next several time steps. The environmental state in future time steps is predicted based on the models constructed above. Then, distributed model predictive control is used to decompose the decision variables to reduce the search space of each computing node (here, each electric vehicle is regarded as a computing node). Then, each electric vehicle uses a genetic algorithm for iterative optimization. After the optimization is completed, the optimal decision is broadcast. This process will be repeatedly executed until Nash equilibrium is reached. After the optimization is completed, the electric vehicle cluster executes the tracking behavior according to the optimization result. When the environment changes, each electric vehicle uses a genetic algorithm and historical data for dynamic iterative optimization. After the optimization is completed, the optimal decision is broadcast. This process will be repeatedly executed until Nash equilibrium is reached.

[0071] The specific evolution process in S3 is as follows:

[0072] S31 Each electric vehicle maintains a population, and the individuals in the population are isomorphic. The dimensions of the individuals are as Figure 3 shown, which are respectively the tracking target number of the electric vehicle at time step k, the maximum driving speed, whether charging is required, the selected charging pile node, until the tracking target number of the electric vehicle at time step k+N-1, the maximum driving speed, whether charging is required, the selected charging pile node and the charging duration;

[0073] S32 Each electric vehicle initializes the population and randomly selects an individual to broadcast.

[0074] S33 For each electric vehicle, the method for calculating the objective function is as follows:

[0075] J(X i (k),{X j≠i (k)},u i [k:k+N-1],{u j≠i [k:k+N-1]},{E[k:k+N-1]}); where the objective function includes two sub-objective functions, namely the effective tracking duration and the vehicle energy consumption. J is the weighted average of the two sub-objective functions of the effective tracking duration and the vehicle energy consumption after being normalized respectively.

[0076] Among them, the formula for the sub-objective function of the effective tracking duration is: where N represents the number of time steps, T represents the number of target objects, and e t is a Boolean function. If within the t time step, for the target object e tInside a circle with the center at point e and radius r, there is an electric vehicle performing a tracking task. t It is 1, otherwise it is 0.

[0077] Among them, the formula for the sub-objective function of vehicle energy consumption is: Where ΔE i,k is the energy consumption of electric vehicle i at time step k, M is the number of driverless vehicles, and ΔE max is the maximum possible energy consumption of all driverless vehicles within one time step (obtained through prior and recorded data), and E max is the energy when each driverless vehicle is fully charged (here it is assumed that all driverless vehicles are homogeneous), and E i,k+N-1 is the energy of the driverless vehicle at time step k + N - 1 (here it is not directly used because the driverless vehicle can be charged, so the energy of the driverless vehicle is not always decreasing).

[0078] For the other parts of the objective function formula, k represents the current time slice number, N represents the next N future time slices to be considered, X i (k) represents some states of electric vehicle i at time step x, such as the position and remaining battery of the electric vehicle, and {X j≠i (k)} represents the states of other electric vehicles in the system at time step k, and u i [k:k + N - 1] represents the task allocation plan of electric vehicle i from time slice k to k + N - 1, and {u j≠i [k:k + N - 1]} represents the task allocation plans of other electric vehicles from time slice k to k + N - 1, and {E[k:k + N - 1]} represents the states of the environment from time slice k to k + N - 1, such as target positions, road network and power grid conditions, etc.;

[0079] Since the states of electric vehicles and the states of the environment at each time step are required for calculating the objective function, for the objective function values of future time slices, the road network traffic flow model, charging pile network performance model, and target trajectory model are used to predict the environmental states.

[0080] For each electric vehicle i, under the condition of fixing {u j≠i [k:k + N - 1]}, the genetic algorithm is used to iteratively optimize u i [k:k + N - 1], and then the optimal individual is recorded and broadcast.

[0081] For each electric vehicle, the Nash equilibrium judgment formula is used to determine whether the Nash equilibrium is reached. If the Nash equilibrium is not reached, return to S34; otherwise, execute the current optimal individual as the task allocation plan. The Nash equilibrium judgment formula is as follows:

[0082]

[0083] where M represents all the intelligent electric vehicles in the system is the optimal individual of electric vehicle i in the l-th iteration is the optimal individual of electric vehicle i in all historical iterations, and ε n is a relatively small positive number and serves as a threshold

[0084] S36 Each electric vehicle saves the current optimal solution to its own database. This database has a capacity limit. When the database is not full, the solution is directly added to the database. Otherwise, the current solution replaces the solution with the worst diversity in the database. The diversity of the database solutions is calculated according to the following formula

[0085]

[0086] where C i,d is the value of the dimension d of solution i, and N C is the number of solutions in the database, and center d is used to represent the center of the database solutions, and dc i is the distance of a certain solution from the center and can be used to characterize the diversity of individuals

[0087] S4 After obtaining the optimal solution, the electric vehicle performs tasks based on the optimal solution. When the electric vehicle is performing tasks, it monitors the vehicle state, target trajectory, road network traffic flow, and the performance of the sensor network in real time. Then, at regular intervals, it calculates the fitness value of the current decision using the updated data. If the fitness value deteriorates, the electric vehicle that recognizes that the solution has degenerated will broadcast this information, and the system will perform dynamic optimization operations

[0088] Such as Figure 2 shown, the specific dynamic optimization operation is as follows

[0089] S41 Increase the population mutation probability

[0090] S42 Shift the previous optimal solution to the left, that is, remove the task assignments for the time steps that have been executed, use the time step corresponding to the current moment as the task assignment for the first time slice of this solution, and then shift it to the left in sequence. Finally, fill the last valid time step into the remaining part. As shown below

[0091] If the solution used by the current electric vehicle i is u i =[a k ,a k+1 ,a k+2 ,a k+3 , and two time steps have been executed, then shifting the solution to the left gives u il =[a k+2 ,a k+3 ,a k+3,a k+3 ;

[0092] S43: Then compare the left-shifted solution with the original optimal solution, retain the better one of the two, and call it the adjusted solution;

[0093] S44: Re-initialize the population, then calculate the fitness value, and replace the individual with the worst fitness value in the population with the adjusted solution;

[0094] S45: Perform the mutation and recombination operations of the genetic algorithm, then calculate the fitness, and select the individuals for the next iteration according to the fitness;

[0095] S46: Randomly select several individuals from the database, and at the same time select the corresponding number of the worst individuals in the current population, then compare the fitness values in turn. If the fitness value of the database individual is better than that in the population, replace the individual in the population with the database individual;

[0096] S47: Perform the Nash equilibrium optimization of S35. If the Nash equilibrium is not reached, repeat the execution: S45 - S46. Otherwise, take the current optimal individual as the current task assignment scheme of the intelligent vehicle, and store this scheme in the database.

[0097] This method comprehensively considers the energy consumption factors of the dual networks (road network and charging pile network) to improve the efficiency and effect of the task assignment of electric vehicles.

[0098] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited by the described embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. An electric vehicle task allocation method considering the energy consumption of dual networks, characterized in that It includes the following: S1 Obtain the trajectory data of each target, obtain the historical traffic flow data of each edge of the road network, and obtain the location information and historical charging efficiency data of each charging pile node in the charging pile network; S2 Based on the trajectory data of the target and the dual-network data, conduct mathematical modeling on the electric vehicle task allocation optimization problem, define a model with the vehicle tracking target number, maximum driving speed, road network edge number, charging pile area number, and charging time as decision variables, and establish an objective function with the effective tracking duration and energy consumption as the objectives; S3 Adopt a distributed dynamic evolution algorithm based on Nash optimization, with the longest effective tracking duration and the lowest vehicle energy consumption as the optimization objectives, to obtain the optimal electric vehicle task allocation method. The electric vehicle task allocation method satisfies the maximum driving speed of the vehicle with the lowest tracking ability, and at the same time allocates the nearest charging pile to the electric vehicle that needs to be charged.

2. The electric vehicle task allocation method according to claim 1, wherein, In S2, based on the trajectory data of the target and the dual-network data, conduct mathematical modeling on the electric vehicle task allocation optimization problem, specifically: Use the LSTM model to model the road network traffic flow, train it with the historical traffic flow data of each edge of the road network, obtain a traffic flow time series prediction model for predicting the traffic flow of each edge of the road network, and predict the traffic flow of each edge of the road network at the next time step; Use the LSTM model to model the charging efficiency of the charging pile network, train it with the historical charging efficiency data, obtain an efficiency time series prediction model for predicting the charging efficiency of the charging pile, and predict the charging efficiency of each charging pile area at the next time step; Based on the Kalman filter, model the target trajectory, train it with the target historical trajectory, obtain a target tracking time series prediction model, and predict the position prediction information of the target at the next time step.

3. The electric vehicle task allocation method according to claim 2, wherein Adopt a distributed dynamic evolution algorithm based on Nash optimization, with the longest effective tracking duration and the lowest vehicle energy consumption as the optimization objectives, to obtain the optimal electric vehicle task allocation method, specifically: S31 For each electric vehicle, calculate the objective function as follows: J(X i (k), {X j≠i (k)}, u i [k:k + N - 1], {u j≠i [k:k + N - 1]}, {E[k:k + N - 1]}); Among them, the objective function includes two sub-objective functions, namely the effective tracking duration and the vehicle energy consumption. J is the weighted average of the two sub-objective functions of the effective tracking duration and the vehicle energy consumption after normalization respectively; k represents the current time slice number, N represents the N future time slices to be considered, and X i (k) represents the state of electric vehicle i at time step k, {X j≠i (k)} represents the states of other electric vehicles in the system at time step k, u i [k:k + N - 1] represents the task allocation plan of electric vehicle i in time slices k to k + N - 1, {u j≠i [k:k + N - 1]} represents the task allocation plans of other electric vehicles in time slices k to k + N - 1, and {E[k:k + N - 1]} represents the state of the environment in time slices k to k + N - 1; The description of the sub-objective function of the effective tracking duration to be minimized is as follows: where N represents the number of time steps, T represents the number of target objects, and e t is a Boolean function. If there is an electric vehicle performing a tracking task within a circle with the target object e t as the center and a radius of r at time step t, then e t is 1, otherwise it is 0; The description of the sub-objective function of the vehicle energy consumption to be minimized is as follows: where ΔE i,k is the energy consumption of electric vehicle i at time step k, M is the number of unmanned vehicles, ΔE max is the maximum possible energy consumption of all unmanned vehicles in one time step, E max The energy of each driverless car when fully charged, E i,k+N-1 is the energy of the unmanned vehicle at time step k. Since the calculation of this objective function involves multiple time steps in the future, it is necessary to bring the decision variables into the above dual-network model to predict the energy consumption; For each electric vehicle i, when fixing {u j≠i [k:k + N - 1]}, the genetic algorithm is used to iteratively optimize u i [k:k + N - 1], and then the optimal individual is recorded and broadcast; S33 Each electric vehicle maintains a population, and the individuals in the population are isomorphic. The dimensions of the individuals are the tracking target number, maximum driving speed, road network edge number, charging pile area number, and charging time of the electric vehicle at time step k; S34 Each electric vehicle initializes the population and randomly selects an individual to broadcast; For each electric vehicle i in S35, when {u j≠i [k:k+N-1]} is fixed, the genetic algorithm is used to iteratively optimize u i [k:k+M-1], and then the optimal individual is recorded and broadcast; S36 Each electric vehicle determines whether it has reached the Nash equilibrium. If it has reached the Nash equilibrium, each electric vehicle stores the optimal solution in the database and executes the task according to this solution. The optimal solution is the electric vehicle task allocation method. Otherwise, it returns to S35 to continue optimization.

4. The electric vehicle task allocation method according to claim 3, wherein, The Nash equilibrium judgment formula: Among them, M represents all the intelligent electric vehicles in the system, is the optimal individual of electric vehicle i in the l-th iteration, is the optimal individual of electric vehicle i in all historical iterations, and ε n is a positive number and serves as a threshold.

5. The electric vehicle task allocation method according to claim 3, wherein It also includes that when the electric vehicle executes the task, it monitors the vehicle state, target trajectory, road network traffic flow situation, and the performance situation of the sensor network in real time. Then, every once in a while, it calculates the fitness value of the current decision using the updated data. If the fitness value is lower than the threshold, the electric vehicle that recognizes that the solution has degenerated will broadcast this information, and then the system will perform dynamic optimization operations.

6. The electric vehicle task allocation method according to claim 5, wherein, The specific dynamic optimization operations are as follows: Increase the mutation probability of the genetic algorithm; Shift the solution to the left: The optimal solution before shifting, that is, remove the task assignments of the executed time steps, take the time step corresponding to the current moment as the task assignment of the first time slice of this scheme, then shift left in turn, and finally fill the last valid time step into the remaining part. Then compare the shifted scheme with the original optimal scheme and retain the better one of the two, which is called the adjusted solution; Re-initialize the population, then calculate the fitness value, and replace the individual with the worst fitness value in the population with the adjusted scheme; Use the database to guide the search: When the fitness is calculated, randomly select several individuals from the database, and at the same time select the corresponding number of the worst individuals in the current population, and then compare the fitness values in turn. If the fitness value of the database individual is better than that in the population, replace the individual in the population with the database individual.

7. The electric vehicle task allocation method according to claim 3, wherein, S36 further includes: When the database capacity is not full, directly add it to the database, otherwise replace the individual with the worst diversity in the database with the current individual.

8. The electric vehicle task allocation method according to claim 7, wherein The diversity calculation formula is as follows: Among them, C i,d is the value of dimension d of individual i, N C is the number of individuals in the database, center d is used to represent the center of the database individuals, dc i is the distance of a certain individual from the center, which is used to characterize the diversity of the individuals.

9. The electric vehicle task allocation method according to claim 3, wherein, The state of the electric vehicle at time step k includes the position and remaining power of the electric vehicle.

10. The electric vehicle task allocation method according to claim 3, wherein, The state of the environment includes the target position, road network state, and charging pile network state.

Images